Brandon Hawkins, Principal Product Manager at Twilio. Twilio is the world’s leading Customer Engagement Platform (CEP), powering direct, personalized, and intelligent customer experiences across every communication channel. Trusted by developers and enterprises in over 180 countries, Twilio helps companies bring AI, data, and security into every touchpoint—from sales and marketing to customer service and growth. With millions of developers building on their platform, Twilio continues to drive innovation in conversational AI and real-time communications, enabling businesses to create magical, scalable interactions with their customers.

Listen to the episode on Spotify, Apple Podcast, Podcast addicts, Castbox. You can also watch this episode on YouTube.

In this episode of the AI Minds Podcast, Brandon Hawkins, Principal Product Manager at Twilio, shares his journey from software engineer to leading AI-driven product innovation at one of the world’s top communication platforms.

Brandon reflects on his transition from early-stage startups to Twilio, highlighting the lessons learned from managing both digital and physical products and the importance of communication and prioritization in large organizations.

He dives into Twilio’s cutting-edge work in conversational AI, explaining how products like Conversation Relay and AI Assistance are enabling real-time, human-like interactions across messaging, voice, and video channels.

The conversation explores the challenges of latency, turn-taking, and voice quality in AI-powered communications—and how Twilio is solving them to help companies deliver personalized customer experiences at scale.

Brandon also discusses the future of voice technology, the ethics of voice cloning, and Twilio’s commitment to trust, transparency, and developer-first innovation as they help shape the future of AI-powered communication.

Show Notes:

00:00 Career Transition: Engineer to Product

04:05 "Curiosity About Corporate Dynamics"

06:58 "Effective Communication via Document Socialization"

12:15 Minimizing Latency in Product Design

13:04 "Voice Agents and User Interaction"

17:08 Balancing Product Design Complexity

21:46 Security-First Approach to Tech Challenges

24:11 AI-Augmented Workflow Strategies

26:01 Future LEGO Adventures Await

More Quotes from Brandon:

Demetrios:

Welcome back to another AI Minds podcast. This is a podcast where we explore the companies of tomorrow being built AI First. I'm your host, Demetrios. And this episode, like every other episode, is brought to you by Deepgram, the number one text to speech and speech the text API on the Internet today, trusted by the world's top enterprises, conversational AI leaders and startups like Spotify, NASA, Citibank and Twilio. Speaking of Twilio, I have the great pleasure of being joined by a principal product manager at Twilio today. Brandon. How you doing, man?

Brandon Hawkins:

Hey, Demetrios, good to see you. Thanks for having me.

Demetrios:

So I want to go over a bit of your story and then I want the jam session to happen on Twilio and all of the conversational AI stuff that you all are doing. But I think you've been in the game, the startup game and then the big company game in tech for a while now. You started as a software engineer and have drifted into product, which I think is a story that happens to quite a few of us now. You're working at Story blocks and then you went into B2B. You went into actual physical products. Can you give me a bit of a story on how the transition from becoming a software engineer to a product person happened?

Brandon Hawkins:

I did. I made the move to the dark side, depending on your perspective, basically out of necessity. I like problem solving, so that was what was attractive to me about software engineering in the first place. And product just sort of represented an opportunity to widen the aperture or the definition of what problems we're solving. So it's customers, it's market. And our first startup, we was Story blocks, like you mentioned, is we were fortunate enough to kind of grow really fast and have strong product market fit. So we quickly got to a place where we had more things coming in to do than we could possibly handle and prioritize.

Demetrios:

And that's always fun.

Brandon Hawkins:

I got the opportunity to cut my teeth there and make a lot of mistakes and learn the dark art of product management. I haven't looked back since now, when.

Demetrios:

You went into physical products and you were doing stuff there. How do you consider these two worlds different? Like between bits and bytes and atoms?

Brandon Hawkins:

They're very different. The financing can be different too from like the cap table structure all the way down to. How we operate, the end user experience. You're a lot more reliant on partners, to produce the thing You've got to design it. you probably have like an E commerce component that's the software digital but ultimately it comes down to like shipping an actual thing and those things have defects. Software.

Brandon Hawkins:

If you have ship a defect, you just fix it. You just push out an update. But for example, this was artifact uprising, high quality print photo books, you screw that up, this is a customer's, it's usually something sentimental. It's a big deal. It's like a wedding photo album. You can't get that back so you gotta, there's lots of cost with that. You gotta make sure it's really right up front.

Demetrios:

Patching a piece of software seems trivial compared to going back and returning a high quality photo album and having to then make a champion out of somebody who is probably pretty pissed off a hundred percent. So then I want to hear about the move from startups to a bigger company, which is Twilio, and how that transition was and has been.

Brandon Hawkins:

I was really curious to kind of see, I always been on the startup side. I've gone through an acquisition, I've seen various stages of startups but I was always curious to see what does it look like on the other side, and sort of learn that skill set and as it turns out it's more of the same. And there's of course some things that are different, but it's always the same stuff. we have more things than we can possibly do. How do we prioritize, how do we listen to customers. But the biggest learning is how to work across a large group of people. Where you have to more lead by influence and get a lot of folks to work together on a common thing.

Demetrios:

This is incredible, this lead by influence piece. I just want to tug on that thread a little bit more because you're not the boss of everyone and you have to almost champion for what you think is right. But you can't really demand to do things like maybe with a startup I would imagine that if you wanted something done, you always have this in your back pocket like I'll just it over the weekend. But at a large company like Twilio, I think that's probably a little harder to try and do.

Brandon Hawkins:

That's 100% true. And there's various flavors. You need to rely on a team and that's the advantage we have too. Is that we do have a lot of folks who are very skilled and people who can lead functions that you don't have the luxury of having even that department at a startup. But the heroics are harder to pull off. And in a small startup you may have a team of five or ten.

Brandon Hawkins:

And everybody knows every detail about context, talking to customers, what's going on. But you have to document that, write it down, make sure folks communication is really important in terms of that.

Demetrios:

That's funny you should say that because I was just talking to a guy, Peter, who wrote a book called on why Software Engineers Should Write More. And it basically boils down to that, like you are so much more effective if you can be a master communicator. And one of the best ways to communicate your thoughts and really materialize them is by writing. And so that was his whole thing.

Brandon Hawkins:

It's one of Twilio's magic values. We come from Jeff Lawson was our founder and he came from the Amazon culture. Very driven into that culture and ours too is write it down. Because to your point, you have to think about it and synthesize it to be able to write clearly and not write a 30 page thing. It's gotta be like a one pager or six pager and people get it quickly.

Demetrios:

I always say too that when you socialize a document around, it's much different than when you're in a meeting and everybody agrees on something. Because when you're in that meeting, maybe you understood one thing, I understood another thing and I forgot one of the things that we were going to write down. And maybe we have the note taker in there and it helps remind me, but it's not quite the same as if we're socializing a document, we're commenting on it, we agree, we align and then boom, it's off to the races. So speaking of communication, I feel like that's a good Segway into what you all are doing at Twilio. It feels like Twilio is kind of made for communication in a way. And I know you have a bit of an origin story. Can you hit me with that real fast?

Brandon Hawkins:

Twilio is the world's leading communication platform. We invented the category of CPAAS communication platform as a service 16 years ago. I think our vision is personalization at scale. And so if you take, we think if you take communications and data, like really Personalized data, first party data, and then blend that with AI. You have a special, you have something special there. And we kind of sit at the middle of that. We've been a developer first company from the beginning.

Brandon Hawkins:

Like I mentioned, Jeff, our founder, he's a developer first product was programmable voice. But Twilio has all the communications channels. But we think we're sitting at a unique place to deliver on that vision now. And the tech is finally there to make these things real. And we're a platform we're not in a foundational model company, but we partner with all the AI companies and play nicely with all of them. And we've got some really cool products that have been cooking up and you guys have really helped us bring to market.

Demetrios:

I want to talk a bit about the products and how you are thinking about from the product perspective, infusing into Twilio different AI capabilities and specifically around upgrading the conversational side of things.

Brandon Hawkins:

So we're calling this whole thing and the markets coalesced around this term conversational AI Cai. You'll hear this a lot more if you don't already, folks listening. It's funny because look, two and a half years ago, ChatGPT didn't exist. And a lot has happened in the time since November of whatever that was. And we saw customers, our customers.

Brandon Hawkins:

Everything starts with the customer, obviously. But our customers about a year ago were trying to stitch together like they have these legacy IVR systems. This is like your call in and press one to speak to an operator. Everyone hates it.

Demetrios:

So painful. we talk about this a lot on this podcast.

Brandon Hawkins:

You hammer zero. You're like give me a human. But we saw folks stitching together, a Deepgram, some other text to speech providers and putting an LLM in the middle. And it was pretty good. it wasn't perfect, but it was pretty good. And a lot has happened. In the past year and a half to make these things better. But we heard from our customers that we want to do this.

Brandon Hawkins:

We think the tech is there. Can you help make it easier for us? We already build communications on Twilio. We think you guys should be playing here. And we said, we think we should too. But latency is a problem. When we talk about. I think all the channels are interesting, Like messaging, voice, email, video.

Brandon Hawkins:

The real time communication channels are where the Problems and therefore the opportunities are the largest because it's hard.

Demetrios:

And what do you mean by real time?

Brandon Hawkins:

You and I are talking, Like over voice. this is a real time. we're doing a lot of things to handle interruptions with each other, to make sure the pacing is human. Like in a texting world, you just send the text. And then there's an async response. So there's a lot of things you have to get right. You can't hide the words in a real time situation.

Brandon Hawkins:

So latency, super critical, interruption, handling super critical. And the voices, they gotta be good too. And everything's gotta be accurate.

Demetrios:

It's funny you mentioned that because a friend of mine who was building a voice agent said that everything seemed fine when he's building. And then he pushes it out and he tries it, he's his own guinea pig. And he starts talking to it, thinking it's going to be great. And he asks the voice agent to do something for him. And next thing there's 30 seconds of silence and he's looking at it on his computer screen and he sees that the agent is trying to take these actions, but he recognizes that as a user you don't get to see that. And so you're looking and you're, that's a horrible experience. You're like it working? Did it freeze? Is something happening? I don't know what's going on a hundred percent.

Brandon Hawkins:

And I think, we take that on the product side really seriously. And that with regards to the things that we can control, we should have that latency as absolutely low as possible to give, developers and customers like your friend Headroom to do those cool things, Because everyone wants to go do a tool call personalization, look up some data, write something back. It does take time. And there's like new UX patterns coming to life that we're putting names to for things that have existed for all of human eternity since we've been speaking. Like if you and I were talking and I had to go get a book or something, I'd be just a second, I'll be right back. That's called an interstitial. And web pages have the loading spinner. But now we have to bring these concepts, the voice world, the communication world.

Demetrios:

That is so funny that concept and many others in this greater conversation of how do we design our conversations and our, our voice Agents to be the most effective. And so another person that I was talking to said that, it's actually not advantageous for me to have the voice agent be really good, because then humans take that for I can say anything and do anything at any pace and in any accent or whatever and mumble my words, and it will pick it up. And it gives this false confidence that if the voice agent is really good, then you can do what you want with it and it will grab anything and take it and be able to use it. Whereas if you make it a little bit worse, it actually gets better results because humans recognize, this thing might not work. And you're used to those traditional systems that are so bad that this is way better than that, but if you're talking to it like it is a traditional system, then you're good.

Brandon Hawkins:

It's really wild to see, kind of the human behavior play out in good and bad ways.

Demetrios:

Yeah.

Brandon Hawkins:

Deepgram has given us a lot of cool tech to make cover up a lot of those words. In terms of basically, it's. If it's garbage in, you're going to get garbage out. And so the accuracy, of transcribing what's happening real time to then send that off to an LLM or be processed by an application, it has to be good. And. But we're also seeing, if you figure out, if a human figures out that they're speaking to an AI agent, it's a bot, their bad behavior comes out too.

Brandon Hawkins:

Thankfully, the bot doesn't care. But people get short, they interrupt it's an interesting to see this play out.

Demetrios:

There's a few other terms that have become more popular, as you were saying, because now we have to put names to these things that we do as humans, but we didn't necessarily need to name them. And one that I heard come up, I don't know if you've seen this one, is callback where if you don't have the right amount of information to get to the next step, if the agent cannot complete the next step's task without hard information, if I'm ordering pizza and I don't give my address, then the agent can't go and order me pizza. It can't send me and deliver me that pizza. So you need to have this stop function or this callback that can before sending it to an LLM and Having it go and do stuff, I still need. I'm in the information gathering phase here.

Brandon Hawkins:

I hadn't heard that term. Obviously that use case and that problem is well known.

Demetrios:

So.

Brandon Hawkins:

I'm learning something here today too. Callback. I love it. there's lots of these.

Demetrios:

Are there other terms? what terms have you heard named turn?

Brandon Hawkins:

Turn taking is a classic one, you and I just naturally know when there's a pause or sometimes there's an interruption, but it's an overlap or we kind of finish each other's thoughts and it's flowing. That is confis it harder than it.

Demetrios:

Than it looks so hard. And actually pulling on that thread. Another one that is hard is that I speak at a certain cadence and I pause. Maybe I have long pauses and then I keep talking. The voice agent and this conversational design, you really want it to be tailored to each person's speaking style. But that is very hard to do also. And so if you're tweaking the knobs of how quickly the agent is going to react or respond, for one person that may be okay, but for another person that speaks a little slower or has longer pauses, that's totally not okay.

Brandon Hawkins:

A hundred percent. And you have put your finger on one of the harder product strategy problems that we face, which is fun too. But do you design products for the ultimate control, like long tail, expose every knob. It's super complicated. Or on the other end of the spectrum, you get this one or two things. And we're very opinionated about that. The way we've approached that is we sort of said we'll give you both and then we'll see what kind of where the boundaries are over time. And Conversation Relay is an example of one of these products that's in Twilio's conversational AI umbrella.

Brandon Hawkins:

And this product was designed fo the ultimate control for ISVs, independent software vendors who want to control that. Maybe even every call, every customer personalize. Or there's a particular bot has a personality. And this is low latency text to speech, speech to text. Deepgram is powering it. And basically customers can bring their own LLM. Twilio is very unopinionated.

Brandon Hawkins:

Just give us basically the stuff we've been doing for 16 years that doesn't make your beer taste better, which is real time communications orchestration. This is our business. we're going to be in it for a long time, you guys. we let our customers focus on what makes their business different and delivering a great user experience. And then the other end of the spectrum we have another product. It's called AI Assistance and this is a whole platform for building AI agents in. Twilio has a great ui but it's very opinionated. Basically you say I just want that outcome.

Brandon Hawkins:

I saw that demo, I want that. I don't know, I don't want to care about LLMs or knowledge or rag or memory or tool calling. Just give me something and let me run. So we kind of are approaching it from both ends and we'll see, where they meet.

Demetrios:

You know what I'm so surprised has not become a thing yet is we've seen that we can voice clone really quickly with very small snippets of text. We can do a really good job voice cloning. Why then can we not have a small snippet of your audio and then personalize the agent right there and know, oh this person seems to talk slower or faster or they seem to have this type of whatever it may be that's going to help you have a better agent experience.

Brandon Hawkins:

It's super interesting. how could you personalize or differentiate? The possibilities are endless. And the truth is it's possible, You can do that. you can listen and do something like that. You could have an input for it from like a UX perspective to sort of set it up. But you could be listening in a conversation and then know, and you could even know your customers preferences before. Like if you remember I mentioned Twilio, we think big picture data AI comms, the data side super important. So we have a CDP and it's required segment, it's now folded into Twilio.

Brandon Hawkins:

But you can collect all customer touch points from marketing, From any sort of form they've ever filled out, any ad they've clicked, any product they've purchased. All these things from disparate systems, pull them together in one place so that you have a great profile like so that golden profile of who a customer is a lot before you even get them on the line, let alone what you can do in real time. So do you tailor the voice like if someone's calling in from. It's a classic example like Southern US has a, very distinct accent draw. Do you tailor the Voice to that or do people get creeped out? I think it's an interesting thing we're going to see play out.

Brandon Hawkins:

Do you want to just talk to Demetrios like your mirror, or that is it.

Demetrios:

That's funny because the idea of potentially it's going to freak some people.

Brandon Hawkins:

Out but just like.

Demetrios:

I guess social media freaked a lot of people out, including myself back in 2008, in 2007 type thing.

Brandon Hawkins:

The way that we're approaching this sort of big picture the sort of trust and guardrails and security type stuff is more conservative. I mean we are a larger company. We have huge enterprises operating on our platform and all down to individual developers. We put security and trust first and we don't want to ship something that going to get folks in trouble or a black box of automation you can't see inside. So those things are really important to us. But the voice cloning stuff you brought up is also super interesting where from an unsolved problem in open domain is how do you do authentication? Or can you over voice now that you can spoof them pretty darn good. And that's an unsolved problem.

Demetrios:

That feels like one that maybe we shouldn't be doing voice authentication anymore and you have to really think through how you're going to be doing it and what you're going to be doing. it's a wild way forward, I would say. But I do want to talk before we jump a bit more about these different products that you had mentioned. I know there's Relay and there's the AI Assistant, which are almost like two sides of the spectrum. Are there any other things that you're particularly stoked about or passionate about right now that you're working on and you can tell me about for sure.

Brandon Hawkins:

Obviously this space is both intellectually interesting and stimulating. Because it's the talk of the town. I was born in 1985, I went through the Internet, pre Internet and then mobile shift. This feels like kind of the next big one.

Demetrios:

Especially with voice. I mean LLMs are great. Image generation is great. But voice man is so wild.

Brandon Hawkins:

What a renaissance. It's happening. The word on the street was voice is dead. Everyone's going to text. That's no one wants to talk anymore. Well, as it turns out, people do want to talk. It's just hard. And when they want to talk, it's the highest value.

Brandon Hawkins:

Most important conversations. Usually it's very mission critical.

Demetrios:

So true.

Brandon Hawkins:

I think the way that we think about it is basically there's like two kind of like pillars and they're virtual agent where you basically automate everything that a human can do and free up the human to do more productive things. And then the other side of the coin is like human in the loop, where if you do have to have a human talking to a customer or having some sort of interaction, they should be super powered by AI, have context, have next best action. There's lots of things that we can do there too. And those are the two big pillars we're we're investing in. And then most importantly is something that's common to all of them like a foundation is this observability, insights, intelligence, observability into the black box. It could be table stakes, stuff like telemetry, was there jitter, was there lag, what was the latency? But all the way up to more interesting things like doing evals on your agents. So that if you change a prompt, is it going to totally change, the expected outcome? What was the sentiment.

Brandon Hawkins:

Of some of the customer interaction? Was a competitor mentioned. There's lots of cool things we can do in terms of like natural language operators. And that's a whole suite of things that Twilio is invested in to basically, our product mission is to basically say we're going to have a suite of things, an end to end solution. And if you just want two or three pieces of it or just one, that's fine too. You can do that, We're not a walled garden. We're a developer first platform. And so we always build these primitives, we call them sort of like the individual LEGO pieces. We'll also have LEGO kits for folks too who say, I just want the space shuttle.

Brandon Hawkins:

Just give me a space shuttle. Great. No problem. You could buy it.

Demetrios:

I have with a daughter who is about 6 years old right now. I know all about LEGO kits. We haven't gotten into the Lego blocks and the creativity galore yet, but I'm hoping that is one direction that we're going to go into. Well, listen man, this has been great talking to you. I appreciate this and I'm excited for everything that you're working on and everything that we're doing together. At Twilio and Deepgram. It's very cool.

Brandon Hawkins:

I really appreciate Demetrios. Appreciate you guys having us on and being good partners for us.